OpenLiDARViewer v0.4.7

A classification, reporting, and honesty release — derive a usable classification for clouds that ship without one, get a clearer client PDF, and read a grade you can trust. Still browser-native, local-first, nothing uploaded.

Highlights

• Classify — derive a classification for unclassified clouds — a one-click action assigns a coarse, ASPRS-aligned class (ground, low/medium/high vegetation, building) to scans that carry no classification channel — raw XYZ, photogrammetry, unclassified LAS. It runs a progressive morphological ground filter, builds a bare-earth surface, measures height-above-ground and local roughness, and labels each point from those geometry cues. It runs off the main thread with live progress, and the result feeds the Classes legend, the colour-by-class mode, GPU class filtering, and exports. It is a heuristic, not a producer's classification — the legend, the toasts, and the export provenance all say so, and it is always tagged "derived" so it can never be mistaken for survey-grade.
• Inspection summary in every PDF report — each template opens with a synthesised findings card: a headline capture type and scale, scannable findings tiered by status, a point-density read, and an explicit list of what the report does not establish. The density-tier call only invokes the USGS QL bands when they actually apply to the capture; vertical accuracy that wasn't measured stays unmeasured.
• Honest preview grading — a surface-quality score computed on a partial or still-streaming sample now reads as provisional (Preview · ~57/100) in both the headline verdict and the detailed score, so a complete, good file no longer shows a low preview number as if it were the file's grade. A streaming COPC that finishes streaming in now earns a real grade instead of being capped at "Preview" forever.
• Colourblind-safe classification palette — a toggle in the Classes panel recolours the classes with an Okabe-Ito categorical palette so ground, vegetation, buildings, and water stay distinguishable under the common colour-vision deficiencies. The class label and count stay on every row, so colour is never the only cue.
• Workflow recorder returns — record a sequence of camera moves and tool actions and replay it on the same scan. A settings popup tunes the file format (readable / compact), the save destination (download or a native save-as dialog), the start/stop shortcut, replay speed (0.5× / 1× / 2× / instant), a pre-record countdown, which action families are captured, and loop replay. It records actions only — never scan data — so a recipient needs the same scan open to replay it.
• Annotation grouping — the Annotations panel and the PDF report open with a one-line summary of the notes: totals, the per-category breakdown (notes / info / warnings / issues), and how many areas they fall across, so a dense set reads at a glance.

Reporting & analysis

• The PDF dataset summary discloses how much of a streaming COPC/EPT scan was actually resident at export time — a "Loaded" row reading, e.g., 4.2M of 15.7M pts (27%) — streaming preview — so a mid-stream report is never read as covering the whole cloud.
• Cross-section profiles export as a real vector chart in the PDF with a vertical-exaggeration disclosure, station chainage, and slope summary; the 3D scene marks each station.
• Metric labels are disambiguated so two different things never share a name: scan scope (full / resident / sampled) reads as "Scan scope" instead of colliding with the measured "Coverage" figure, and the readiness card's measured fraction reads as "Measured coverage" instead of clashing with the composite "DTM quality" score.
• The Classes legend discloses that streaming per-class counts are a running tally over decoded nodes, not full-file totals.
• The deep "All metrics" breakdown is collapsed by default; the decisive chips carry the headline.
• Below-grade surfaces are described by fitness-for-use ("usable for inspection and measurement, not yet for terrain-product export") rather than a deliverable / non-deliverable binary.

Viewer

• A freshly opened scan frames to its full extent (extent-aware fit), centred and a little larger, so it reads on first open without a manual zoom.
• The Rendering panel shows the point size in pixels and defaults to adaptive sizing.
• The Classes panel widened so class names don't truncate.
• An honesty cue on the Dataset Intelligence card: a quiet coloured dot marks each row's qualitative tier; terrain complexity stays neutral (it is descriptive, not a quality), and a missing signal is muted to match the —.

Correctness & honesty fixes

• The ASPRS class table is complete through codes 0–22 (Reserved, Overlap, Overhead structure, Ignored ground, Snow, Temporal exclusion), in both the point readout and the colour palettes.
• Loading a second scan adds it as a separate layer again, rather than replacing the first — restoring multi-scan loading.
• Empty files are rejected with a clear message instead of opening a blank, unframable scene.
• Reprojection never ships non-finite coordinates — a transform outside the target projection's valid area fails honestly and leaves the source untouched.
• NAD83 against the WGS84-coincident datum family is flagged as an identity shift (~1–2 m), alongside the existing NAD27 and GDA94↔GDA2020 caveats.
• A measured area reads the same in the PDF report as on screen.
• Unknown point density shows — rather than a fabricated "Sparse".
• Streaming sources close their reader on detach; the colour-recompute timer is cleared on teardown.
• A worker-fallback diagnostic no longer claims "fell back to main thread" before the safety check that can refuse the fallback — the log now states only what actually happened.

Under the hood

• A performance pass across the load, streaming, and analysis paths to keep the browser responsive on large clouds.
• The unit test suite is split into four coverage-complete buckets (test:unit / test:terrain / test:ui / test:slow) that always union to the whole suite, so it can run in parallel.
• Packaging now verifies every release archive (unzip -t) and fails rather than shipping a truncated or corrupt zip, and a bundle-budget check guards the shipped build against silent size growth.

Measurement, derived classification, and the floor-plan preview are for visual inspection and research, not survey-grade use. Treat any output as survey-grade only after validating it against survey-grade data and procedures.

See CHANGELOG.md for the full list.

Deploy

Static files — host anywhere (GitHub Pages, Netlify, any CDN). The deploy zip extracts with web-safe permissions (644 files / 755 directories) and carries index.html plus assets/ at the archive root.